It is UB to construct out-of-bounds pointers.
Signed-off-by: Brian Anderson <andersrb@gmail.com>

Previous code creates aliasing mutable slices.
Signed-off-by: Brian Anderson <andersrb@gmail.com>

Stop advancing src_ptr after the empty terminator group (remaining==0).
The old loop always did ptr::add(MEMCMP_GROUP_SIZE) after every group;
after the final empty padded group that steps past one-past-the-end of
the allocation (UB under Rust pointer rules / Miri), same class as tikv#7751.
When len is a multiple of the group size the encoder still correctly
emits a full data group (padding 0) then one empty terminator group;
only the post-terminator advance is removed.
Also adds miri_soundness_mem_comparable_encode_all_naive which, under
Miri, fails on pre-fix code and checks equality with encode_all after
the fix.

Stop advancing src_ptr after the empty terminator group (remaining==0).
The old loop always did ptr::add(MEMCMP_GROUP_SIZE) after every group;
after the final empty padded group that steps past one-past-the-end of
the allocation (UB under Rust pointer rules / Miri), same class as tikv#7751.
When len is a multiple of the group size the encoder still correctly
emits a full data group (padding 0) then one empty terminator group;
only the post-terminator advance is removed.
Also adds miri_soundness_mem_comparable_encode_all_naive which, under
Miri, fails on pre-fix code and checks equality with encode_all after
the fix.
Signed-off-by: Paulo Cabral Sanz <paulo@railway.app>

Stop advancing src_ptr after the empty terminator group (remaining==0).
The old loop always did ptr::add(MEMCMP_GROUP_SIZE) after every group;
after the final empty padded group that steps past one-past-the-end of
the allocation (UB under Rust pointer rules / Miri), same class as tikv#7751.
When len is a multiple of the group size the encoder still correctly
emits a full data group (padding 0) then one empty terminator group;
only the post-terminator advance is removed.
Also adds miri_soundness_mem_comparable_encode_all_naive which, under
Miri, fails on pre-fix code and checks equality with encode_all after
the fix.
Signed-off-by: Paulo Cabral Sanz <paulosanz@poli.ufrj.br>